How to measure whether AI actually names you
Ask the assistants your customers' questions on a fixed schedule and count how often you are named. There is no report to open: Google merges AI Overview and AI Mode clicks into overall Search traffic and does not break them out. Twenty questions across four platforms once a month is an afternoon, and it is the only honest measurement available.
Why there is no report to open
The first thing to accept is that the obvious tool does not answer the question. Google includes AI Overview and AI Mode clicks in overall Search traffic and reports them within the Web search type. No setting splits them out. An owner who checks Search Console for AI performance is reading a number that contains the answer without revealing it.
The second thing to accept is that clicks are the wrong unit anyway. With 82 percent of consumers reading AI generated summaries and 23 percent deciding on the summary alone, a mention that never produces a click can still produce a customer. Counting clicks would systematically undercount the channel.
The measurement, in four steps
One, fix the questions. Write down fifteen to twenty questions in the exact words a customer would use, including the town. Never change the list, because a changing list measures nothing.
Two, fix the platforms. Test where the traffic actually is. ChatGPT carries 74.78 percent of AI referrals, Gemini 11.56 percent, Perplexity 7.23 percent and Copilot 3.51 percent, so four platforms cover about 97 percent of the channel.
Three, record three things per question: were you named at all, in which position among the businesses listed, and which source the assistant cited. The third is the useful one, because it tells you which page to improve.
Four, repeat monthly and write the numbers in one sheet. Twenty questions across four platforms is eighty checks, which is an afternoon.
| Column | What goes in it | Why it matters |
|---|---|---|
| Question | The customer's exact wording, with the town | A fixed list makes months comparable |
| Platform | ChatGPT, Gemini, Perplexity, Copilot | Those four are about 97 percent of AI referrals |
| Named | Yes or no | The headline number |
| Position | Where among the businesses listed | Being fourth is not being first |
| Source cited | The URL the assistant credited | Tells you which page to improve |
Reading the result honestly
Two traps wait at the end. The first is treating one run as a measurement. Assistants are not deterministic, and the same question can return different businesses an hour later. Only the trend across several months means anything.
The second is measuring the wrong success. A mention rate of 20 percent on twenty questions is not a failure if those four mentions come from the questions with buying intent. Weight the list before you start: five questions that end in a purchase are worth more than fifteen that end in curiosity. Cross check against the referral rows in your analytics, since chatgpt.com and perplexity.ai do appear there, and a rising mention count with flat referrals usually means your landing page is the problem rather than your visibility.
Questions and answers
Can I not just use a rank tracker?
How many questions are enough?
Why does the same question give different answers?
What counts as a good result?
Do AI referrals show up in ordinary analytics?
Sources
- Google Search Central, AI features in Google Search That AI Overview clicks are not reported separately in Search Console
- SE Ranking, AI traffic research study, 18 June 2026 101,574 websites; platform shares used to choose what to test
- BrightLocal, Local Consumer Review Survey 2026 1,002 consumers; reading of AI summaries
